Abstract 15653: Optimizing the Screening Algorithm for Critical Congenital Heart Disease: A Data-Driven Approach
Introduction: Screening for critical congenital heart disease (CCHD) using pulse oximetry is performed with various algorithms in the United States, with none of these based on robust data, including the one endorsed by the AHA and other organizations.
Hypothesis: Existing algorithms will have differing results for CCHD screening, and the test characteristics of these algorithms can be improved through modifications.
Methods: After collecting de-identified, patient-level pulse oximetry values from existing national and international screening programs, we compared the commonly used algorithms in the US: the American Academy of Pediatrics (AAP) algorithm (endorsed by the AHA), the New Jersey (NJ) algorithm (slight modification of the AAP algorithm), and the Tennessee (TN) algorithm (screening first on the foot only). We then considered alternate algorithms by modifying the pulse oximetry values that would be considered a pass, fail, or retest result in the existing algorithms. We compared the algorithms according to sensitivity (Se), false positive rate (FP), and area under the curve (AUC).
Results: There were 75,748 screenings, with 57 true positives. Performance characteristics of the 3 common algorithms are summarized in the Table. The NJ algorithm had the greatest Se but also the highest FP, and the TN algorithm had the lowest Se but also the lowest FP. Modifications of these algorithms yielded 2632 potential algorithms with Se ranging from 28% to 91%, FP from 0.05% to 60%, and AUC from 0.64 to 0.82. Of these, 224 algorithms met our optimal algorithm criteria of Se >50%, FP 0.75.
Conclusions: Current algorithms for CCHD screening vary in their test performance characteristics, and there are opportunities to modify these algorithms to improve test performance. Before identifying an optimal algorithm, further work is needed to collect more data (including more robust data on false negatives) and to compare the cost and ease of use of potential algorithms.
Author Disclosures: M. Oster: None. C. Caglayan: None. R. Simeone: None. P. Keskinocak: None. T. Ayer: None.
- © 2015 by American Heart Association, Inc.